Recently, Carter et al1 reported the prevalence of p16INK4a gene deletions and their prognostic value in a cohort of 45 children with initial acute lymphoblastic leukemia (ALL). Using a real-time quantitative duplex polymerase chain reaction (PCR) assay, the prevalence of homozygous and hemizygous p16INK4a deletions was 25% and 13%, respectively. We performed a similar assay on 125 samples from children with first relapse of ALL and would like to comment on their data.
To date, real-time PCR is not widely used for detecting hemizygous deletions in genomic DNA because concentration differences as small as 0.5-fold have to be detected. These differences become even smaller when nonmalignant cells are present in the sample (up to 25% contaminating cells in the specimens investigated by Carter et al). To allow the unequivocal detection of hemizygous deletions, the assay must be extremely reproducible. In our hands, despite thorough optimization of a real-time PCR assay analyzing target and control genes in separate tubes, the intra- and interassay reproducibility is insufficient. Carter et al have not assessed the reliability of their assay. They defined the ratios for hemizygous deletion as ranging from 0.4 to 0.8, but no experimental rationale to support these values is given. Their mixing experiments were carried out on DNA from cell lines with ap16INK4a homozygous deletion and wild type, respectively, but did not include a cell line hemizygous for ap16INK4a deletion. Values between 0.25 and 0.5, as well as between 0.75 and 1, are difficult to interpret, and there is neither experimental nor theoretical background for the values the authors defined to discriminate hemizygously deleted from homozygously deleted and wild-type genotypes. Allowing for no more than 25% contaminating normal cells, ratios of samples with hemizygous deletions should range from 0.5 to 0.75, while homozygous deletions should be no more than 0.25, and wild-type genotypes should always be 1. Values interpreted as hemizygosity could theoretically represent polyclonality, aneuploidy, a combination of these, or true hemizygosity. Considering methodological limitations and the difficulties in data interpretation, we suggest using the real-time PCR technique only for the detection of homozygous deletion. For a unequivocal diagnosis of hemizygous deletion, fluorescence in situ hybridization (FISH) should be used.
The prognostic importance of p16INK4a deletions in childhood ALL is still controversial. Using multivariate Cox regression analysis for the probability of relapse-free survival on 45 patients, Carter et al found hemizygous and homozygous deletion ofp16INK4a to be an independent predictor of poor outcome. These results support a previous report from their laboratory and from Heyman et al.2 In contrast, 2 studies not cited by the authors on a larger number of patients did not find a prognostic impact of p16INK4a deletions.3 4The number of patients Carter et al investigated is too low for the use of multivariate Cox regression analysis with so many variables, and neither did they show any odds ratios of the confounding factors included in their analyses nor did they present data about the association that might exist betweenp16INK4a loss and relevant prognostic factors. In particular, it would be of interest whetherp16INK4a deletion remains of independent prognostic significance within the subgroups of patients with T-cell or B-precursor ALL, respectively.
We have analyzed 125 samples from children with first relapse of ALL for p16INK4a deletions using a real-time quantitative PCR assay with minor differences from the method of Carter et al. The children were treated according to trials ALL-REZ BFM 90 and 96 of the Berlin-Frankfurt-Münster Relapse Study Group.5 The prevalence of homozygousp16INK4a deletions was found to be 35% (44 of 125). Due to considerations mentioned above, hemizygous deletions were not assessed. Homozygous p16INK4a deletions showed a highly significant association with the most important adverse prognostic factors for relapsed ALL,5 T cell immunophenotype and early occurrence of relapse (less than 6 months after cessation of front-line therapy). Despite this strong correlation with adverse prognostic factors, no independent prognostic impact ofp16INK4a deletion could be found in our cohort. The probability of event-free survival at 5 years forp16INK4a deletion and nondeletion was 0.46 ± 0.08 and 0.43 ± 0.07, respectively.
The prevalence of the homozygous p16INK4adeletion in relapsed ALL is not much higher than at first diagnosis, a fact one should expect from a prognostically nonsignificant alteration. With the absence of any prognostic impact of the p16INK4a deletion in relapsed ALL, it is unlikely that a prognostic influence of the p16INK4adeletion in initial ALL is independent from other confounding variables. To finally clarify this issue, which has generated controversy for nearly 6 years, we agree with Carter et al that prospective evaluation on a larger number of patients with appropriate methods should be carried out.
In conclusion, real-time PCR is a convenient method to rapidly analyze homozygous deletions in a large number of patient samples. Detection of hemizygous deletions with this method is questionable and should be evaluated by FISH. Our data on p16INK4a gene deletion analysis in 125 children with first relapse of ALL do not support the hypothesis presented by Carter et al regarding an independent negative prognostic influence in childhood ALL.
p16INK4A gene deletion in pediatric acute lymphoblastic leukemia
Einsiedel et al comment on the prognostic value of thep16INK4A gene deletion in pediatric acute lymphoblastic leukemia, with particular reference to the methodology used in our recent publication.1-1 The real-time polymerase chain reaction (PCR) was developed in our laboratory to detect deletion of the p16INK4A exon 2 gene since the Southern blotting method used in our previous studies does not allow accurate quantitation of gene deletion. The novel method described by us is performed in a multiplex format where the p16 test gene is amplified in the same tube as the reference gene, β-actin. In order to establish that the technique has the capacity to detect gene deletion in specimens containing normal cells, we conducted mixing experiments using DNA from 2 cell lines, one showing homozygous deletion of p16 (D/D) and the other being wild type (G/G). This test was performed multiple times using independent samples of DNA mixtures and reproducibly yielded a linear graph with a correlation coefficient of 0.9687 to 0.9742 between the input ratio of mixed DNA and the experimentally determined ratio of p16. The reproducibility of the assay was further examined by repeatedly analyzing patient specimens representing the 3 p16 genotypes G/G, G/D, and D/D. These specimens were measured in 6 replicas each, showing ratios of 0.91 ± 0.03 for the G/G specimen, 0.69 ± 0.08 for the G/D specimen, and 0.13 ± 0.02 for the D/D specimen (mean ± SD). The same specimens were measured in 2 additional independent experiments (in duplicates each), and the values (mean ± SD) from the 3 tests were 0.95 ± 0.13, 0.63 ± 0.07, and 0.07 ± 0.05 for the G/G, G/D, and D/D specimens, respectively. Taken together, these results indicate that the interclass correlation coefficient measuring the reproducibility of measurements for a given genotype is 0.909.
We elected to conduct the mixing experiment using 2 cell lines, rather than a cell line showing hemizygous deletions. Apart from the fact that we do not have access to a line showing this particular feature, such a cell line may contain submicroscopic lesions in p16 not detected by cytogenetics. The mixing experiment provides much more information as it allows titration over the entire range, from 0 to 100 percent. This is of critical importance as it was necessary to focus on the range between 0% and 25% to assess the suitability of the test for patient specimens containing normal cells. The ratios to determine the p16 genotype of the patient specimens was based on the reproducibility of the assay (see above), and we opted to use conservative values of 0.4 and 0.8. Most importantly, the patient specimens showed a clear triphasic distribution, consistent with discrete populations having G/G, G/D, or D/D alleles.
Einsiedel et al were not able to establish a technique as accurate and reproducible as ours, which may be due to the instrument used or/and to the fact that test and reference genes were not measured in a multiplex reaction. Our results on many standard calibration curves showed that the conditions optimized for multiplexing pass the test for the comparative efficiency test, which means that they should allow detection and comparison of the test gene and reference gene in separate tubes. Although we did not expect to achieve the required accuracy to conduct the analysis in separate tubes, we verified whether our conditions would be suitable. Indeed, the results confirmed this to be the case, but as expected, the accuracy in repeat tests is not as high as by the multiplexing method.
As stated in our paper, we intended to confirm the status of the G/D specimens by using an independent technique. Due to lack of material, it was not possible to conduct fluorescence in situ hybridization studies. Similarly we refer to the controversy regarding p16deletion as a prognostic factor in pediatric ALL. Rather than quoting many individual publications, we referenced the reviews by Drexler et al1-2 and by Tsihlias et al,1-3 which contain all relevant publications. Moreover, as stated in our paper, we agree with Einsiedel et al regarding the need for a larger study to assess the significance of p16 deletion as a prognostic marker, and such a study is currently in progress in our laboratory.
In order to determine the frequency of p16 deletion at relapse, we studied patients from whom we obtained diagnosis and relapse bone marrow specimens.1-4 The data showed the rate to be much higher at the time of relapse, which is in agreement with a study by Maloney et al.1-5
We are intrigued by the motivation underlying the statement “The number of patients Carter et al investigated is too low for the use of multivariate Cox regression analysis with so many variables, and neither did they show any odds ratios of the confounding factors included in their analyses nor did they present data about the association that might exist between p16INKAloss and relevant prognostic factors.” The first part is not substantiated; we would strongly recommend that personal beliefs about the theoretical validity of an analysis might usefully be accompanied by an explanation of the theoretical basis of those beliefs. There are plenty of unique failure times to ensure that the number of risk sets underpinning the generation of the partial log likelihood in this case is adequate to permit the number of parameters we use in our analysis to be estimated. Obviously the data set is relatively small, but that is why we quote confidence intervals and why we state that further investigation is essential. As is almost always the case, severe space limitations prevented us from including (1) the results of formal model checking (completeness of linear predictor; analysis of Martingale residuals; checks of leverage and influence), which showed that our primary models fitted well, or (2) the associations between baseline potentially confounding covariates and outcome and between the covariates and p16INK4A loss. We happen to agree that such data ought to be provided, and they were in fact included in earlier longer versions of the paper, but the reality is that requirements for radical abridgment ultimately meant that they had to be removed. It is of relevance to note that, had these results been reported, it would not in any way have changed the conclusions of the paper. In essence, we agree with much of what Einsiedel et al write, and in particular are delighted that they support our call for further research. There is no question that the differing results from the various studies to which we and they refer are intriguing and need to be properly understood. In particular, it would be of interest whetherp16INK4A deletion remains of independent prognostic significance within the subgroups of patients with T-cell or B-precursor ALL, respectively.
References
This work was supported by a grant of Deutsche Kinderkrebsstiftung to H.G.E.
